Artificial neural networks based models for the multiply excited switched reluctance motor

The flux per phase of multiply excited switched reluctance motors (SRM) deviates from the flux per phase in the singly excited case. Likewise, the net torque deviates from the value obtained by superposition of the results from the singly excited case. This is due to mutual flux interactions between the excited phases and nonlinearities in the system. This paper describes a simplified approach, based on feedforward (FF) artificial neural networks (ANN), to account for these mutual interactions and for the estimation of torque when simultaneously exciting two phases. The proposed technique requires a small measured data set and involves simple calculations. This technique can be applied in simulations as well as in real-time implementations for online phase-flux and torque computations. A description is included of an implementation on the Texas Instruments (TI) TMS320C6701 DSP. The paper also describes a scheme for measurements and quantification of the mutual flux interaction between the phases.